Are you modeling carryover and lagged effects for promotions and seasonality effects? You Should.

In Marketing Mix Modeling (MMM), almost all vendors adstock transform their media variables. This is done because the ad displayed today has a lingering impact into the future and more airing of ad sometimes stops having a incremental effect (the saturation).

However when it comes to assessing the impact of a promotional effect or holiday, no carryover or lagged effect is taken into consideration.

In most MMMs, holidays and events are treated like switches. Example:

Black Friday = 1
All other days = 0

This form of encoding is also known as ‘Dummy encoding’.

But this is to simplistic and wrong. Because human behavior doesn’t follow calendar boundaries.

📌 Events also behave like media

Let’s take an example:

A Black Friday deal creates anticipation -> browsing -> delayed purchase

A World Cup or Super Bowl doesn’t just spike viewership on match day alone. It builds momentum, conversation and residual engagement.

Hence all of these are not a ‘point-in-time’ effect.

What is perplexing to me is, we model media with memory but we model events as memoryless !!

This is clearly not the right way to go about.

We should ideally model carryover + lag for events and holidays too.

In fact we too didn’t until Nov 2024. Through our breakthrough RBF paper we solved this problem. More on this shortly.

📌 Traditional MMM with Dummy Encoding – The wrong approach

When you use dummy variables:

– You assume instant impact
– You assume zero pre-effect
– You assume zero post-effect

But in Reality:

– There is build-up effect (anticipation)
– There is peak or crescendo(event window)
– There is decay (after-effect)

A single binary variable cannot capture this shape.

So what happens?

▪️You over-credit the event week
▪️You miss pre-event demand creation
▪️You misattribute post-event sales to media or base

📌 How we started measuring effect of Events accurately through RBF.

Radial Basis Functions (RBFs) are built for exactly the above problem.

At their core, they model effects as distance-based smooth curves around a center point.

Instead of: Event = 1 or 0

We get a continuous influence curve around the event:

If you don’t model event’s carryover or lag properly:

You will inflate media ROI during event period
You will make wrong budget decisions post-event

And worse, You will think your model is “working”. Because hey R squared will still look good.

📌 The Success Story

Since Jan 2025, we have incorporated this technique for all our clients. Roughly 50 models. In all our models we saw a remarkable improvement in MMM metrics.
The clients also expressed a greater satisfaction in having measured the effect more effectively.

We also had multiple MMM analysts from different companies write to us sayin that RBF technique is really a gamechanger and it improved their models too. We open sourced our research and code for anybody to
try. (Link in comments).